A new study published on arXiv explores the impact of AI-generated pre-annotations on human disagreement in video labeling tasks. Researchers found that while AI assistance significantly reduced annotation time by 23.11% and increased inter-annotator consistency, it did not substantially collapse semantic-label entropy. The study suggests that AI pre-annotations primarily function as editable temporal scaffolds, though subtle semantic anchoring effects remain a possibility. The research introduces a novel audit framework and anonymized interaction traces to further investigate AI's influence on human disagreement in annotation processes. AI
IMPACT Investigates how AI tools influence human judgment in data labeling, potentially affecting dataset quality and diversity.
RANK_REASON Academic paper detailing a study on AI's impact on human annotation. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- CatalyzeX Code Finder for Papers
- DagsHub
- Gotit.pub
- Hugging Face
- Juan Gutiérrez
- Pre-Annotations
- ScienceCast
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →